Files
nvfp4-megamoe-kernel/vllm
biondizzle 29f8b8c174 fix: load lm_head.weight in outer model before forwarding to inner
lm_head lives on DeepseekV4ForCausalLM, not DeepseekV4Model. The inner
load_weights silently drops it (not in params_dict). Extract it in the
outer loader, load it directly, then forward the rest to the inner model.
2026-05-15 02:17:16 +00:00
..